/*
* __las_page_instantiate --
* Instantiate lookaside update records in a recently read page.
*/
static int
__las_page_instantiate(WT_SESSION_IMPL *session,
WT_REF *ref, uint32_t read_id, const uint8_t *addr, size_t addr_size)
{
WT_CURSOR *cursor;
WT_CURSOR_BTREE cbt;
WT_DECL_ITEM(current_key);
WT_DECL_ITEM(las_addr);
WT_DECL_ITEM(las_key);
WT_DECL_ITEM(las_value);
WT_DECL_RET;
WT_PAGE *page;
WT_UPDATE *first_upd, *last_upd, *upd;
size_t incr, total_incr;
uint64_t current_recno, las_counter, las_txnid, recno, upd_txnid;
uint32_t las_id, upd_size, session_flags;
int exact;
const uint8_t *p;
cursor = NULL;
page = ref->page;
first_upd = last_upd = upd = NULL;
total_incr = 0;
current_recno = recno = WT_RECNO_OOB;
session_flags = 0; /* [-Werror=maybe-uninitialized] */
__wt_btcur_init(session, &cbt);
__wt_btcur_open(&cbt);
WT_ERR(__wt_scr_alloc(session, 0, ¤t_key));
WT_ERR(__wt_scr_alloc(session, 0, &las_addr));
WT_ERR(__wt_scr_alloc(session, 0, &las_key));
WT_ERR(__wt_scr_alloc(session, 0, &las_value));
/* Open a lookaside table cursor. */
WT_ERR(__wt_las_cursor(session, &cursor, &session_flags));
/*
* The lookaside records are in key and update order, that is, there
* will be a set of in-order updates for a key, then another set of
* in-order updates for a subsequent key. We process all of the updates
* for a key and then insert those updates into the page, then all the
* updates for the next key, and so on.
*
* Search for the block's unique prefix, stepping through any matching
* records.
*/
las_addr->data = addr;
las_addr->size = addr_size;
las_key->size = 0;
cursor->set_key(
cursor, read_id, las_addr, (uint64_t)0, (uint32_t)0, las_key);
if ((ret = cursor->search_near(cursor, &exact)) == 0 && exact < 0)
ret = cursor->next(cursor);
for (; ret == 0; ret = cursor->next(cursor)) {
WT_ERR(cursor->get_key(cursor,
&las_id, las_addr, &las_counter, &las_txnid, las_key));
/*
* Confirm the search using the unique prefix; if not a match,
* we're done searching for records for this page.
*/
if (las_id != read_id ||
las_addr->size != addr_size ||
memcmp(las_addr->data, addr, addr_size) != 0)
break;
/*
* If the on-page value has become globally visible, this record
* is no longer needed.
*/
if (__wt_txn_visible_all(session, las_txnid))
continue;
/* Allocate the WT_UPDATE structure. */
WT_ERR(cursor->get_value(
cursor, &upd_txnid, &upd_size, las_value));
WT_ERR(__wt_update_alloc(session,
(upd_size == WT_UPDATE_DELETED_VALUE) ? NULL : las_value,
&upd, &incr));
total_incr += incr;
upd->txnid = upd_txnid;
switch (page->type) {
case WT_PAGE_COL_FIX:
case WT_PAGE_COL_VAR:
p = las_key->data;
WT_ERR(__wt_vunpack_uint(&p, 0, &recno));
if (current_recno == recno)
break;
WT_ASSERT(session, current_recno < recno);
if (first_upd != NULL) {
WT_ERR(__col_instantiate(session,
current_recno, ref, &cbt, first_upd));
first_upd = NULL;
}
current_recno = recno;
break;
case WT_PAGE_ROW_LEAF:
if (current_key->size == las_key->size &&
memcmp(current_key->data,
las_key->data, las_key->size) == 0)
break;
if (first_upd != NULL) {
WT_ERR(__row_instantiate(session,
current_key, ref, &cbt, first_upd));
first_upd = NULL;
}
WT_ERR(__wt_buf_set(session,
current_key, las_key->data, las_key->size));
break;
WT_ILLEGAL_VALUE_ERR(session);
}
/* Append the latest update to the list. */
if (first_upd == NULL)
first_upd = last_upd = upd;
else {
last_upd->next = upd;
last_upd = upd;
}
upd = NULL;
}
WT_ERR_NOTFOUND_OK(ret);
/* Insert the last set of updates, if any. */
if (first_upd != NULL)
switch (page->type) {
case WT_PAGE_COL_FIX:
case WT_PAGE_COL_VAR:
WT_ERR(__col_instantiate(session,
current_recno, ref, &cbt, first_upd));
first_upd = NULL;
break;
case WT_PAGE_ROW_LEAF:
WT_ERR(__row_instantiate(session,
current_key, ref, &cbt, first_upd));
first_upd = NULL;
break;
WT_ILLEGAL_VALUE_ERR(session);
}
/* Discard the cursor. */
WT_ERR(__wt_las_cursor_close(session, &cursor, session_flags));
if (total_incr != 0) {
__wt_cache_page_inmem_incr(session, page, total_incr);
/*
* We've modified/dirtied the page, but that's not necessary and
* if we keep the page clean, it's easier to evict. We leave the
* lookaside table updates in place, so if we evict this page
* without dirtying it, any future instantiation of it will find
* the records it needs. If the page is dirtied before eviction,
* then we'll write any needed lookaside table records for the
* new location of the page.
*/
__wt_page_modify_clear(session, page);
}
err: WT_TRET(__wt_las_cursor_close(session, &cursor, session_flags));
WT_TRET(__wt_btcur_close(&cbt, 1));
/*
* On error, upd points to a single unlinked WT_UPDATE structure,
* first_upd points to a list.
*/
if (upd != NULL)
__wt_free(session, upd);
if (first_upd != NULL)
__wt_free_update_list(session, first_upd);
__wt_scr_free(session, ¤t_key);
__wt_scr_free(session, &las_addr);
__wt_scr_free(session, &las_key);
__wt_scr_free(session, &las_value);
return (ret);
}
/*
* __wt_curfile_create --
* Open a cursor for a given btree handle.
*/
int
__wt_curfile_create(WT_SESSION_IMPL *session,
WT_CURSOR *owner, const char *cfg[], int bulk, int bitmap,
WT_CURSOR **cursorp)
{
WT_CURSOR_STATIC_INIT(iface,
__wt_cursor_get_key, /* get-key */
__wt_cursor_get_value, /* get-value */
__wt_cursor_set_key, /* set-key */
__wt_cursor_set_value, /* set-value */
__curfile_compare, /* compare */
__curfile_equals, /* equals */
__curfile_next, /* next */
__curfile_prev, /* prev */
__curfile_reset, /* reset */
__curfile_search, /* search */
__curfile_search_near, /* search-near */
__curfile_insert, /* insert */
__curfile_update, /* update */
__curfile_remove, /* remove */
__wt_cursor_reconfigure, /* reconfigure */
__curfile_close); /* close */
WT_BTREE *btree;
WT_CONFIG_ITEM cval;
WT_CURSOR *cursor;
WT_CURSOR_BTREE *cbt;
WT_CURSOR_BULK *cbulk;
WT_DECL_RET;
size_t csize;
WT_STATIC_ASSERT(offsetof(WT_CURSOR_BTREE, iface) == 0);
cbt = NULL;
btree = S2BT(session);
WT_ASSERT(session, btree != NULL);
csize = bulk ? sizeof(WT_CURSOR_BULK) : sizeof(WT_CURSOR_BTREE);
WT_RET(__wt_calloc(session, 1, csize, &cbt));
cursor = &cbt->iface;
*cursor = iface;
cursor->session = &session->iface;
cursor->internal_uri = btree->dhandle->name;
cursor->key_format = btree->key_format;
cursor->value_format = btree->value_format;
cbt->btree = btree;
if (bulk) {
F_SET(cursor, WT_CURSTD_BULK);
cbulk = (WT_CURSOR_BULK *)cbt;
/* Optionally skip the validation of each bulk-loaded key. */
WT_ERR(__wt_config_gets_def(
session, cfg, "skip_sort_check", 0, &cval));
WT_ERR(__wt_curbulk_init(
session, cbulk, bitmap, cval.val == 0 ? 0 : 1));
}
/*
* random_retrieval
* Random retrieval cursors only support next, reset and close.
*/
WT_ERR(__wt_config_gets_def(session, cfg, "next_random", 0, &cval));
if (cval.val != 0) {
__wt_cursor_set_notsup(cursor);
cursor->next = __curfile_next_random;
cursor->reset = __curfile_reset;
}
/* Underlying btree initialization. */
__wt_btcur_open(cbt);
/* __wt_cursor_init is last so we don't have to clean up on error. */
WT_ERR(__wt_cursor_init(
cursor, cursor->internal_uri, owner, cfg, cursorp));
WT_STAT_FAST_CONN_INCR(session, cursor_create);
WT_STAT_FAST_DATA_INCR(session, cursor_create);
if (0) {
err: __wt_free(session, cbt);
}
return (ret);
}
/*
* __curfile_create --
* Open a cursor for a given btree handle.
*/
static int
__curfile_create(WT_SESSION_IMPL *session,
WT_CURSOR *owner, const char *cfg[], bool bulk, bool bitmap,
WT_CURSOR **cursorp)
{
WT_CURSOR_STATIC_INIT(iface,
__wt_cursor_get_key, /* get-key */
__wt_cursor_get_value, /* get-value */
__wt_cursor_set_key, /* set-key */
__wt_cursor_set_value, /* set-value */
__curfile_compare, /* compare */
__curfile_equals, /* equals */
__curfile_next, /* next */
__curfile_prev, /* prev */
__curfile_reset, /* reset */
__curfile_search, /* search */
__curfile_search_near, /* search-near */
__curfile_insert, /* insert */
__wt_cursor_modify_notsup, /* modify */
__curfile_update, /* update */
__curfile_remove, /* remove */
__curfile_reserve, /* reserve */
__wt_cursor_reconfigure, /* reconfigure */
__curfile_close); /* close */
WT_BTREE *btree;
WT_CONFIG_ITEM cval;
WT_CURSOR *cursor;
WT_CURSOR_BTREE *cbt;
WT_CURSOR_BULK *cbulk;
WT_DECL_RET;
size_t csize;
WT_STATIC_ASSERT(offsetof(WT_CURSOR_BTREE, iface) == 0);
cbt = NULL;
btree = S2BT(session);
WT_ASSERT(session, btree != NULL);
csize = bulk ? sizeof(WT_CURSOR_BULK) : sizeof(WT_CURSOR_BTREE);
WT_RET(__wt_calloc(session, 1, csize, &cbt));
cursor = &cbt->iface;
*cursor = iface;
cursor->session = &session->iface;
cursor->internal_uri = btree->dhandle->name;
cursor->key_format = btree->key_format;
cursor->value_format = btree->value_format;
cbt->btree = btree;
/*
* Increment the data-source's in-use counter; done now because closing
* the cursor will decrement it, and all failure paths from here close
* the cursor.
*/
__wt_cursor_dhandle_incr_use(session);
if (session->dhandle->checkpoint != NULL)
F_SET(cbt, WT_CBT_NO_TXN);
if (bulk) {
F_SET(cursor, WT_CURSTD_BULK);
cbulk = (WT_CURSOR_BULK *)cbt;
/* Optionally skip the validation of each bulk-loaded key. */
WT_ERR(__wt_config_gets_def(
session, cfg, "skip_sort_check", 0, &cval));
WT_ERR(__wt_curbulk_init(
session, cbulk, bitmap, cval.val == 0 ? 0 : 1));
}
/*
* Random retrieval, row-store only.
* Random retrieval cursors support a limited set of methods.
*/
WT_ERR(__wt_config_gets_def(session, cfg, "next_random", 0, &cval));
if (cval.val != 0) {
if (WT_CURSOR_RECNO(cursor))
WT_ERR_MSG(session, ENOTSUP,
"next_random configuration not supported for "
"column-store objects");
__wt_cursor_set_notsup(cursor);
cursor->next = __wt_curfile_next_random;
cursor->reset = __curfile_reset;
WT_ERR(__wt_config_gets_def(
session, cfg, "next_random_sample_size", 0, &cval));
if (cval.val != 0)
cbt->next_random_sample_size = (u_int)cval.val;
}
/* Underlying btree initialization. */
__wt_btcur_open(cbt);
/*
* WT_CURSOR.modify supported on 'u' value formats, but the fast-path
* through the btree code requires log file format changes, it's not
* available in all versions.
*/
if (WT_STREQ(cursor->value_format, "u") &&
S2C(session)->compat_major >= WT_LOG_V2)
cursor->modify = __curfile_modify;
WT_ERR(__wt_cursor_init(
cursor, cursor->internal_uri, owner, cfg, cursorp));
WT_STAT_CONN_INCR(session, cursor_create);
WT_STAT_DATA_INCR(session, cursor_create);
if (0) {
err: /*
* Our caller expects to release the data handle if we fail.
* Disconnect it from the cursor before closing.
*/
if (session->dhandle != NULL)
__wt_cursor_dhandle_decr_use(session);
cbt->btree = NULL;
WT_TRET(__curfile_close(cursor));
*cursorp = NULL;
}
return (ret);
}